So I was reading this and it got me thinking. Lens flare is just light refracting on camera lenses, right? So it should solely depend on one thing: how the sun is occluded from your viewpoint. Which is already computed, since you have to draw the sun and the occluding object anyway.It's still not 'full' testing (e.g., actually computing how much of the flare is occluded), because that would be too expensive.
Wouldn't it be as simple as one or two matrices or image manipulating steps to create a believable lens flare?